A finite volume method for two,dimensional

transonic potential flow through

turbomachinery blade rows
J. V, Soulis*

To predict inviscid transonic flow through turbomachinery blade rows, the exact
transonic potential flow equation is solved on a mesh Constructed from small area
elements. A transformation is introduced through which distorted squares of the
physical plane are mapped into computational squares. Two sets of overlapping
elements are used; while the thermodynamic properties are calculated at the primary
element centres, the flux balance is established on the secondary elements. For
transonic flows an artificial compressibility term (upwind density gradient) is added
to density in order to produce the desired directional bias in the hyperbolic region,
while the entropy does not increase across mass conservative shock jump regions.
Comparisons with experiments and with other numerical and analytical solutions
for various turbomachinery configurations show that this approach is comparatively
accurate, reliable, and fast.

Keywords: turbines, compressible flow, finite volume methods

In recent years significant advances have been made finite-element method of setting an irregular compu-
in computational fluid dynamics applied to tur- tational grid to fit a particular flow-field problem and
bomachinery. Many approaches have been taken to to modify it with ease, if necessary, and it makes use
the general problem of turbomachinery flow. The of the simple analysis of finite-difference methods.
purpose of this investigation is to develop an efficient Computations of 2-D, blade to blade inviscid,
numerical algorithm for the calculation of 2-D poten- transonic flow can be formulated, inter alia, in terms
tial incompressible and compressible, transonic invis- of either the Euler equation or the fully potential
cid flows through turbomachinery blade rows. equations. To solve Euler's equations Gopalakrish-
A major difficulty in solving the exact potential nan 2 applied the numerical scheme proposed by Mac-
flow equation lies in the construction of sufficiently Cormack comprising two steps (predictor and correc-
accurate discrete approximations to the boundary con- tor) while Denton s used a time marching method
ditions. A mesh may be constructed from small cells based on a finite-volume technique.
(volumes) which are packed around the tur- Full potential, transonic equation solutions
bomachinery blades and cover the whole blade to were reported by Caspar et al 4 where a general non-
blade area (sl) surface, shown in Fig 1. Then a transfor- orthogonal mesh was used; Acay and Ecer s used the
mation is introduced through which the cells of the
physical plane are mapped into computational
squares. The numerical scheme is constructed using
two sets of overlapping cells. The thermodynamic
properties are calculated at the centres of the primary
cells and then the flux balance is established on the
secondary cells. For supersonic regions of the flow-
field, a modified density is used instead of the physical
density as suggested by Hafez et al 1.
The scheme used can be regarded as a combina-
tion of the numerical techniques of finite-elements
and finite-differences; 'finite-volume' seems an
appropriate name. The main features of the numerical *4L

scheme are that it preserves the practicability of the

* Whittle Laboratory, Department of Engineedng, University of

Carnbddge, Cambridge, England. Presently 8t Fluid Mechonics/Hy-
drsulics Division, Civil Engineering Department, Demol(fition Uni-
versity of Thrace, Xanthi, Greece
Received on 27 August 1982 and accepted for publication on
5 September 1983 Fig I $1 and $2 stream surfaces

Int. J. Heat & Fluid Flow 0142-727X/83/040229-09503.00 © 1983 Butterworth & Co (Publishers) Ltd 229
J. V, Soulis

finite element method; Habashi et al 6 used a finite- and, as will be proved later:
element approach with a scheme based upon approxi-
mate mapping of a blade into a near circle; Farrell = p - ~ °_P.PAs
08
and Adamczyck v also used conformal transformation
to generate the grid around the cascade blades.
~ a s =-Sp~
u Ax
v
+-Sp~, AU (8)
Os q q
with
Flow equations
The flow will be assumed to be two-dimensional,
inviscid,potentialand steady,but with no restriction
/z = max I00 1 k

to subsonic flow. Thus the governing flow equations No assumption will be made about the geometry of
for the physical domain, where the Cartesian coordin- the blade row. The problem must be d o s e d with a
ate system is introduced, are, complete specification of the boundary conditions.
Complex turbomachinery geometries present
a(#u) + a(#v) = o (I) difficulty when attempting to prescribe the geometry
ax ay on an orthogonal computational mesh, such as the
Cartesian form. Difflculties can arise because the grid
a~ a~ points do not adequately describe the blade boun-
u =-- v =-- (2)
Ox oF daries. A body fitted non-orthogonal local coordinate
system was used to avoid the problems associated with
q = (Us+ Vs)l/s (3) a conventional grid system. Thus the flow field is
divided into multiple blocks to accommodate all
To 1 = constant (4) boundary surface discontinuities and to provide good
boundary fitting. This finite-element approach is well
Pol = constant (5) adapted to turbomachinery flows for a number of
reasons. A wide range of finite elements is possible
for adequate description of the surface boundaries
(6) and m~nimal additional computational code logic is
\Poll
required for changing a particular element to a more
qS ~ 1/'r-1 sophisticated one. There is an automatic treatment of
. = 2C- o;) (7) the various forms of the applied boundary conditions.

Notation U,V Contravariant velocity components along

the ~, W directions(m/s)
a Local speed of sound (m/s) U, f) Cartesian velocity components along the
atj Transformation weighting coefficients x, y directions(m/s)
Cp Coefficient of specific heat at constant X, y Cartesian coordinate position along the
pressure (kJ/kg K) axial and pitchwise directions(m)
Cv Coemcient of specific heat at constant /31 Inlet flow angle measured from the axial
volume (kJ/kg K) direction (degrees)
Cx Axial blade chord (m) y Specific heat ratio
G Matrix of geometrical coefficients Artificialcompressibility
g~t Elements of the G matrix Transformed coordinate directions(m)
g" Elements of the G -z matrix
P Density (kg/m z)
H Transformation matrix from Cartesian to
local ~, ~7 system
hU. Mass fluxes along the ~, ~1 directions(kg/s)
h Determinant of the H matrix hVpJ
i,j Logical indices of mesh variables along ~7, Potential (mS/s)
directions
k Constant (Eq (8)) Subscripts
M Local Mach number is Isentropic
N, Shape functions o Stagnation value
P Pressure (N/m s ) 1, 2 Upstream, downstream
P Pitch (m)
Q Total velocity in computational plane Superscripts
(m/s)
q Total velocity (m/s) T Transpose matrix
B Gas constant (kJ/kg K) - 1 Inverse matrix
$ Streamwise direction (m) * Sonic condition
T Temperature (K) Modified value

230 Vol 4, No 4, December 1983

 2-D transonic potential flows in blade rows

For an iterative approach these boundary conditions The local coordinates lie in the range -1 s 6 d 1, -1 s
can be applied explicitly, ie from the previous iter- r) s 1 so that the vertices of the square are at 6 = *l r) =
ation, and/or implicitly ie initially set forth. Moreover *1.
a single analysis can generate several finite-difference Let H be the transformation matrix from the
techniques which are dependent on the finite-element physical system to the computational local coordinate
selection. system :
\ lax !!!!I
General transformation
(11)
The essence of this numerical scheme is that distorted
squares in the physical domain will be separately
mapped to squares in the computational domain by
independent transformations from Cartesian x, y to Let G be the product matrix HTH where T denotes
local 5; n coordinates (Fig 2). The distorted squares the transpose matrix with elements gll, g12, gal, g,,.
are packed around the blades (Fig 6) and cover the Let G-’ be the inverse of matrix G with elements g’i,
blade to blade area. A primary cell, which is a distor- g12, g21, g22.
ted square in the physical plane or just a square in The U, V velocity components in the computa-
the computational plane, is formed by eight nodes tional domain are given as:
(Fig 3) while a secondary cell interlocks with nine rim
primary cells. Thus the nodes of the secondary cell
are the centres of the surrounding primary cells. Shifts (12)
of plus or minus one yield the centroid position of
the next secondary cell.
If xi, y, are the Cartesian coordinates of a cell,
while the y v velocity components in the physical
then any coordinates of the cell can be expressed as:
domain are:
8
x= c zv*r*
i-1
Y= ?
i-1
where Ni are the shape functions associated with the
Ni!A (9)
[:I =#I
(13)

The continuity equation in the local system is:

cell nodes. The shape functions are defined in terms
of a local non-orthogonal coordinate system 5,~ as: -a(dU+abWo
- (14)
N _(I-5)(I-7))(--57-I) at 871
I- where h is the determinant of matrix H.
4

NaJ1+s)(I-9)(~-17-I)
Numerical formulation
4
The physical interpretation of the quantities phV, phV
N3=(I+5)(1+)1)(z+9-I) is that these are the mass fluxes across the faces t=
4 constant, 7 = constant of the secondary cell, as shown
in Fig 3. Derivatives of x at the primary or secondary
N _(l-r)u+~)(-k+7)-1) finite-volume centres 6 = 0, VJ= 0 are calculated from:
4-
4
00)
N5=(1-52)(1-7)
\a51i,, 2
2 (15)
ax = %+1,j
& = Cl- T2)(l+ 5)
2
()
-
a77 i,j
- Xf-lj
2

N _ (I - Z”)(I + 7) Primary cell

7- Secondary Cell
2 centre cmtn?
I /
N _ (I - V2)(I - 4)
s-
2

7
4 3 -I ,I
01 111

5
b&Y) 11 (Es?)
8 6 -I,0 t
0 ( It0
Y
li-2jl 1
I 5 2 -1,-I I-I
o-i
c?B
x PHYSICAL DOMAIN COMPUTATIONAL OOMAIN
PHYSICAL DOMAIN COMPUTATIONAL DOMAIN

Fig 2 Dhtorted squares mapped into squares Fig 3 A secondary cell interlocks nine primary cells

ht. J. Heat & Fluid Flow 231

J. V. Soulis

Similar expressions hold for any of the quantities where:

y, dp, phU, phV.
In a sense this discrete approximation is a direct = p - ~ a-ePt~s
08
~As =~8~pa~+~8,p an
finite-element method in which the finite-element
representation of the solution in terms of nodal values (9.0)
and prescribed functions within the elements is with:
directly substituted into the governing equations to
obtain the nodal value equations. The formula for the 1
Q = (U~+ V~)~/~
local flux balance can he written down by applying
it to the secondary cells centred at i, j. The discretisa- (21)
tion is made possible by using partial derivatives intro-
k being typically 0.8 and upwind differencing is used
duced above. Thus:
in evaluating 8~p A~, 8~p Av/as follows
(phU),.j+~ - (phU),,j_,
~ p A~=Pld-Pi./-2 if Uid>O
+ (phV),+~,j - (phV),_la = 0 (16)
8~p A~:=p~j+~-p~.j if Uij~<O
The approximation of the local velocity components (o.2)
is given by: 6.p AT = p~,j- p~-~,j if V~j > 0
6.7p AT =Pt+~,j-Pt,j if Vt,j<~O
This scheme maintains an upwind influence in the
differentiation for supersonic points anywhere in the
mesh for all possible orientations of the velocity vec-
tor. When the solution is converged, the application
(17) of the mass conservation and energy equation to the
v,,j = g~ [ *,,J+, -24' ,,,-, ] series of interlocking finite-volumes will yield zero
net-flux into each element so that overall mass and
energy are conserved. From the above it is evident that
,, r~,,+,.,-4,,_,,,-[ the transonic flow analysis applies when the shock
+g"J L 2 J waves are weak (M ~<1.3), ie entropy does not increase
Thus the discrete approximation of the governing flow across mass conservative shock jump regions.
equation in the local coordinate system is:
al~-],j-i + a2dP~-l,j+l

+ a3~. ~+l,j+l + a4([~ i+l.j_ I

(18)
+ a s <D~-~,1 + a6(:I)~,~+~
+ a7 (I) ~+2,t + as~ ~,t-~
+ a9~I>,d = 0
i~l 1
where the a~, a~ . . . . a9 are functions of the p, h, g~ at
appropriate nodal positions 8. i,j-t ii [ i i*l

x i-I,j
Transonic flow
PHYSICAL DOMAIN COMPUTATIONAL DOMAIN
The above analysis completes the definition of the
discrete approximation scheme for main stream nodes Fig 4 B o u n d a ~ cell for the second computational
in subsonic flow. Numerical formulation of the tran- row (i = 3, j= 1, JM)
sonic problem, however, requires the construction of
an appropriate set of differential equations. Problems
encountered in the development of an efficient
numerical algorithm for the calculation of transonic
flows are:
• the equations governing the flow motion are highly
non-linear;
• the flow is a mixed type with both subsonic and
supersonic flows co-existing with the boundaries
between them unknown; and
• the solution will in general contain shock waves,
whose position is also unknown. i+ld
When the artificial compressibility method is
applied (Appendix I), the continuity equation i,jq ij 4J+,
becomes: x

o(~hU) +a(~hV) = 0 (19) Fig 5 Boundary cell for first computational row (i =
a~ an 2, j = l , JM)

232 Vol 4, No 4, December 1983


Boundary conditions
Suction and pressure surface
When the continuity equation in the local coordinate
system is applied, the fluxes are taken across the faces
of the secondary cell which is bounded on one face
by the body surface. Fig 5 shows the boundary cell
for the computational grid points lying on the first
row. Half the cell is hidden inside the solid surface.
When the flux balance is established on the secondary
cell i, j the flux through the solid body of the blade
must be zero, ie phVtj = 0. The flux along the ~ direc-
tion at the i 4-1, j grid point lying in the flowfield is
taken over twice the area of the flux along the
direction.
Due to the complexity of the finite-volume it
is evident that the next computational grid points
lying on the second row will still have one face
bounded by the blade solid surface, shown in Fig. 4.
Similar analyses are applied for the upper 'pressure'
blade surface.
Periodic boundaries and Kutta-Joukowski condition
It w i l l be assumed that the flow is identical in each
blade passage of the machine so that only one passage
needs to be considered (Fig 6). The boundary condi-
tions at the inlet lines A B (A'B') are subject to the
condition of periodicity specified by the equality of
the complete velocity vector and the tangential veloc-
ity components at corresponding periodic points of
the two lines. It follows that values of • at correspond-
ing points can differ only by a constant amount A~a,,
thus,
~ A'B' = ~ AB 4- A(I~in
Interior element (cell)
ory elements
Pm'ibo~ i ¢ / ~ - bound~m
i.lMPI ,,~/V'J/)('~/~XJ ~ -" i . a r E M l ~ ~
~.lM 4 / j / y X X / ' / ~
/ ] / ) f X X / r / ~/1/1/ JLE i=JTE~R'~
~ X ~ bOundOries PeriodE
N becomes increasingly supersonic, the scheme is
~=4/r/7=3
;=~ ~=z
A i=; D
i= 2
j= I
x Mach numbers in the subsonic regions of the blade
Fig 6 Cascad~ g e o m e t ~ and ~ntte-oolume elements
(cells)
Int. J. Heat & Fluid Flow
2-D transonic potential flows in blade rows
where A~j. determines the mean pitchwise velocity
at the inlet region of the cascade. On the lines
C D ( C ' D ' ) formed by the extensions from the blade
surfaces in the outlet region, the boundary conditions
are also subject to a condition of periodicity specified
by the equality of the magnitude of the velocity vector
and of the tangential velocity components at corres-
ponding periodic points. Thus:
4JC'D, = ~CD + A4~o~t (24)
The A~out must be found through an iterative pro-
cedure so as to satisfy the Kutta condition on the
blades. This condition demands equal pressures to be
obtained at the control points of the elements adjacent
to the trailing edges (Fig. 6). The A~out is calculated
from:
A~out = ~ XMPI,.[TEP1-- ~2,J"TEPI (~5)
where:
I]VIPI,]TEPI = ~) IA4PI,JTEA41
+ q2,]TE ((~IMPI,jTEPI__~IMPI,jTEM1) (26)
q ~MP~,]'rE
with ] T E P 1 = ] T E + 1 and I T E M 1 = J T E - 1 ( ] T E
is the last computational grid point on the blade
surface). When converged, then q2J'rE qlMPI.JTE.
=
For non-lifting cases and for nozzle flow calcu-
lations, the current version of the numerical code skips
the whole potential jump evaluation process.
Numerical resolution
A numerical algorithm has been presented for tran-
sonic flow calculations in turbomachinery blade rows.
(23) Iterations were continued until the average change
of axial velocity between successive iterations
dropped below 0.001%. Run time per point per iter-
ation was 2.2x10-4s on IBM370-165 computer.
Total number of iterations required for convergence
was about 300 depending upon the geometry com-
plexity and type of flow (subsonic-supersonic). Total
CPU time for 7 x 50 grid points mesh was about 30
seconds on the above computer.
For subsonic flows, the central difference rep-
resentations are accurate to second order, while in
supersonic regions the difference approximation is a
combination of the second order central difference
(subsonic regions) with the first order accurate upwind
approximation resulting from the artificial compressi-
bility addition for supersonic regions. As the flow
increasingly retarded in the upwind direction. The
choice of factor k in the equation:
j=JM
strongly affects the accuracy of the solution but not
the stability of the scheme. Errors associated with
values of the factor k higher than 0.8 are increased
suction sudace while for supersonic regions, on the
same blade sudace, the Mach number is significantly
decreased.
233
J. V. Soulis

The most common source of inaccuracy arises
at a blunt leading edge where the relatively coarse
grid cannot resolve adequately the rapid changes in
the flow properties. A large number of grid points
must be used to obtain an accurate solution around a
blunt leading edge.
Many other forms of discrete approximation of
the partial differential equations could have been
selected by altering the order of the finite elements
involved. With the current code, sufficient flexibility
exists for special regions of mesh where resolution is
otherwise inadequate due to steep gradients in the
dependent variable.
The numerical scheme can be used with any
grid formation which need not be uniformly spaced
in either coordinate direction. The grid is not restric-
ted to that shown. Many mesh generation techniques
can be adapted provided that grid periodicity is
maintained.
Applications
Subsonic flow in a NASA turbine starer cascade
The first compressible flow test is an untwisted nozzle,
tested by Whitney et al 9 in a parallel walled annulus.
The problem selected is flow at the mean diameter
section operating at the design mass flow rate with
inlet M* = 0.231 parallel t o t h e axial direction. The
experimental exit flow angle, which is also the total
flow turning angle, was found to be -67.25 °. Fig 7
shows the cascade geometry and the calculated con-
tours of equal Mach* number in the flowfield. Com-
parison between theory and experiment is shown in
Fig 8. Although the agreement is satisfactory, some
differences appear on the blade pressure surface at
about 30% of the blade chord downstream of the
leading edge. Calculated flow angle was found to be
-66.30 °.

1.0

0.8
O

0.6 I
x
®
e
xg
0.4
ollg x

,coo ~ Uo
0.2 x Finite volume
• Time marching

Fig 7 Calculated contours of equal Mach * number

0.0 I I I
3 4 5 6 7 8 9 I0 II 12
(a-0.3340; b-0.4460; c-0.5850; d-0.7440; e-0.8569; Cx ,cm
f--O.7630; g--O.8180) for NASA nozzle turbine cascade
at M* = -0.231 Fig. 9 Comparison between finite-volume and time.
marching predictions for blade surface Mach * number
distributions for NASA nozzle turbine cascade at
1.0 M*, = O.e31

x ==x ~ xex°~
0.8 k
X
XHK-
0.6 X X
×lg
0.4 Y
x~, x~
OX X
,_~xv X
0.2 x Predicted

0.0 I I
3 4 5 6 7 8 9 I0 II 12
Cx, crn

Fig 8 Comparison between measured and finite-

oolume predictions for blade surface Mach * number Fig 10 Calculated contours of equal Mach* (a-
distributions for NASA nozzle turbine cascade at 0.664; b-0.816; c-0.923; d-l.00O; e-1.097)number
M* = 0.23] for Hobson's second impulse cascade at M* --0.61

234 Vol 4, No 4, December 1983

 2-D transonic potential flows in blade rows

Comparison between the finite-volume and Comparison between the calculated results and
Denton's method3 is shown in Fig 9; the agreement the analytical data, obtained from Hobson's hodo-
is good. In both methods an identical computational graph design, is shown in Fig 11. The agreement is
grid was used; this includes cusps at the leading and remarkably good. The Mach number distribution pre-
trailing edges of the blade to avoid discontinuities in serves the symmetry of the blade geometry. The calcu-
geometry. lated exit flow angle is -46.37 °. Calculated results are
also compared with those taken from a transonic
streamline curvature method written by Katsanis It.
Transonic flow in Hobson's second impulse cascade
Although the agreement between the two solutions is
An impulse blade was designed and tested by Hob- good, the maximum differences, 8M* =0.07, occur
son 1° in the transonic wind tunnel at the Whittle on the first and last 30% of the blade suction surface
Laboratory. F i g 10 shows the Hobson's second (Fig 12).
impulse cascade geometry and the calculated contours
of equal M*. The blade is symmetrical about the
mid-chord. The critical Mach number at the cascade Transonic now in a turbine rotor of Rotor-Derby
profile
inlet, M*, is 0.61 which is also the cascade outlet
critical Mach number. The inlet flow direction is The method has been used to determine the flow field
46.10 ° and it is equal to the exit flow direction in within a transonic cascade of Rotor-Derby (RD) pro-
absolute terms. file turbine rotor blades, shown in Fig 13 with the
calculated contours of equal M. Experimental data
1.2 has been provided from commissioning tests TMcarried
out in the transonic wind tunnel at the Whittle
1.0 Laboratory and from similar tests performed on the
same blade profile at the DFVLR-AVA Gottingen,
Germany.
0.8

*~ 0.6

0.4

0.2 --+ Finite volume'

- o Hobson's hodogroph solution-
0.0 I l I I I
0 ' I0 20 3 0 4 0 5 0 6 0 70 80 90 I00
Cx,%

Fig 11 Comparison between Hobson's hodograph

design and the finite-volume solution of the blade
surface Mach* number distribution for Hobson's
second impulse cascade at M* = 0.61

1.2
+
\
L
1.0 v (9

L + J~
(9 •~ (9
4. _
0.8
:f
(gk
r ~-~ ~"~
\
0.6 !:~b ei, 1 ' c../(3 4"
~+~

0.4 I I I "
+ Finite volume I
• Kotsonie' etreomline
O.Z ¢urvoture solution
I I I I I
0.0 l I I I I
0 I0 20 30 40 50 60 70 80 90 I 0 0
cx,%

Fig 12 Comparison between Katsanis" streamline Fig 13 Calculated contours of equal Mach number
curvature solution and the finite-volume solution of (a-0.045; b-0.916; c-1.149; d--1.213; o-1.262; f-
the blade surface Mach* number distribution for Hob- 1.319) at iffi0* (~l =56.0*) and M=ffil.194 for RD
son's second impulse cascade at M ] = 0.61 turbine rotor cascade

Int. J. Heat & Fluid Flow 235

J. V. Soulis

Experimental local Math number distributions ~l = 56'00 MI = 0.460

on the blade surfaces, taken at DFVLR-AVA, with an /32 = -66.70 M2 = 0.916
inlet flow angle/3~ of 56.0* for a range of isentropic
exit flow Math numbers of M2 = 0.85, 1.0, 1.12, and measured results give:
1.181 are shown in Fig 14. Calculated local Math
/3z = 56.0* MI = 0.449
number distributions at the same inlet flow angle for
M~, = 0.916, 0.996, 1.094, 1.194 are shown in Fig 15. ~2=-66.8 ° M2=0.914
All applied exit flow conditions have produced Comparison between theory and experiment
choked flow through the cascade. (DFVLR-AVA) at the highest exit flow condition,
Direct comparison of the calculated and M21, = 1.194, is shown in Fig 17. The comparison
measured (Whittle Laboratory) results at M2t, = 0.916
worsens at about 20-30% of blade axial chord length
and/31 = 56.0 ° is shown in Fig 16. The overall Math on the suction surface where there is high flow turn-
number distribution is in good agreement with the
ing. Measured results show a shock-wave at 75% of
experiment. Calculated results give: the blade axial chord while Schlieren pictures taken
at the Whittle Laboratory show a shock-boundary
1.4 o layer interaction which leads to local separation of
the boundary layer with slow re-attachment.
mE rqrq
1.2 m~ ee 1.4
!

1.0 ! A 1.2
Xxkx ~
| x ~,, ,~ ~ 6o,~
x
0.8 1.0 ~ ~ o %~
_! | XO

0.6' I =E 0 . 8 x~ ×
x !~<lr
l M21o 0.6, x x
0.4 x
| <~ I. 181
| 1.120 0.4 x=
0.2 ; n_ =
• 1.0 . ~x x Predicted
'XXx, x~ ~ x ~ x! X ~ MZ=0.916
.= '=0.85 0.2
o.o I 1 o Meosured
o IO 20 30 40 5o so 70 80 90 I00 /~fZi,=(~'914
C=,% 0.0
0 tO 20 30 40 50 60 70 80 90 I00
Cx ,%
Fig 14 Experimental DFVLR,AVA, blade surface
Mach number distribution at i =0 ° ( ~ =56.0 °) and Fig 16 Comparison between measured (Whittle
M~t,=0.85, 1.0, 1.12, 1.181 for RD turbine rotor Laboratory) and predicted blade surface Mach num-
cascade ber distribution at i = O* (f~l = 56.0*) and M~=0.916
for RD turbine rotor cascade
1.4
1.4 ov ~ x x x

1.2 Xx xp>X i u~ e~x¢

1.2
X ~ ix 0 ®
P® X

1.0 w 1.0 Ix ~
=0 = A x
X

~Z 0 . 8 0.8 o~Y~ x
~Z
I ® A
I X x
0.6 0.6'
| _111" x

i it i ! x~ x ~
0.4 0.4
.mll~ x Predicted
O 1.194- ~XXx> x~ X <~ x 5x~ MZ=I.194
'llllllllllll I I I III-- • 1.094 0.2
0.2 • Measured
• 0;996 ¢ iMzi,,=l'lm--
• 0.916- 0.0
0.0 I
0 I0 20 50 40 50 60 70 80 90 I00
0 I0 20 30 40 50 60 70 80 90 I00 c=,%
CI , %
Fig 17 Comparison between measured (DFVLR-
Fig 15 Predicted blade surface Mach number distri- A VA) and predicted blade surface Mach number dis-
bution at i = 0 ° (131 =56.0*) and M~=0.916, 0.996, trtbutton at i = O*(gj = 56.0*) and M2= l.194 for RD
1.094, 1.194 for RD turbine rotor cascade turbine rotor cascade

236 Vol 4, No 4, D e c e m b e r 1983

 2-D transonic potential flows in blade rows

Conclusions
A finite-difference scheme resulting from a second
order finite element shape function has been applied
to arbitrary turbomachinery geometries using a body
fitted non-orthogonal local coordinate system. It is
able to calculate 2-D potential flows in the incom-
pressible, subsonic, transonic and shock-free super-
sonic flow speed range. The main advantage of the
method is the ease of treatment of exact blade surface
boundary conditions. This has been achieved by
transformation of the usually highly curved tur-
bomachinery blade surfaces to coordinate lines of the
Al=Pemlb¢ I
For 2-D full potential equations the corresponding
"viscous" transonic flow equation can be written~:
--~[(a~-q~)dP.+a*dP..]f~&s(a~-q~)dP. (I.1)
where s and n are coordinates in the local stream and
normal directions, and the right-hand side equals:
p ~,[u[u2u, +2uvv, + vtvv)
; L-;I q= . ,

computational domain. Stability in supersonic regions v [u2uz +2uvu¥ +vavv)

is maintained by the addition of an artificial compress- o= .,] (1.2)
ibility term. This greatly simplifies the numerieal pro-
cedure and effectively allows a mixed elliptic hyper- because:
bolic problem to be solved by an elliptic solution u v
procedure. The method has been applied to a variety ~,=, = - ( ~ - ) , +-(~-)v (I.3)
q q
of turbomachinery blade profiles in order to validate
its potentialities. and:
Numerical results indicate that the finite- u2 _ 2uv v2
volume code is an accurate, reliable and fast method ~- = q--'~¢P=~+ q"'~¢-v + q"'~¢w (I.4)
for predicting the irrotational inviscid, transonic, 2-D
flow through all types of turbomaehinery blades. The first bracket term of Eq (I.2) is written as:
P-[. a ~ ~[u[u~u~+2uvv~+v2vvt ]
References L;t q" LJ
(I.5)
1 H,dez M. M., Murmann E. M. and South J. C. Artificial
compressibility methods for numerical solution of transonic an equivalent expression is given by:
full potential equation. AIAA Paper No 78-1148July 1978
2 Gopalakrislman S. and Bozzola R. A. Numerical technique [(1-~)uAS[q(--~)(uu.+vv.)
for the calculation of transonic flow in turbomachinery cas-
cades. ASME Paper No 71-GT-42, 1971
3 Denton J. D. A time marching method for two and three
dimensional blade to blade flows. Aeronautical Research
Council R&M, 3775, 1975
4 Caspar J. R., Hobbs D. E., and Davies, R. L. The calculation OFf
of two dimensional compressible potential flow in cascades
using finite area techniques. AIAA 17th Aerospace New
Orleans 1979
[ (: :)]
IzuAs P, + Pv
X
(I.7)
5 Acay H. V. and Ecer A. Transonic flow computations in
cascades using a flnite-element method. ASME Paper No. with:
81-GT-4 1981
6 Habashi W. G., Dueek E. G. and Kenny D. F. Finite-element
approach to compressor blade to blade cascade analysis.
AIAA Journal, ]ul{i 1979 17(7) because:
7 Farrell C. and Adamczyck J. Full potential solution of tran-
sonic quasi-3-D flow through a cascade using artificial com-
pressibility. ASME, No. 81-GT-70 1981 p~=--~(uu=+vv,) and Pv=--~(uuv+v%)
8 Soulis J. V. Calculation of transonic potential flow through (I.9)
turbomachinery blade rows. Ph.D. Thesis Cambridge Uni-
versity 1982 thus:
9 Whitney, W. J. et al Cold air investigation of a turbine for (~up, As), (I.10)
high temperature engine application. I-turbine design and
overall engine performance. NASA TN D-3751, 1967 Similarly the second member of the bracket yields:
10 Hobson D. E. Shock-free transonic flow in turbomachinery
cascades. PKD. Thesis, Cambridge University 1974 (ttvp, As) v (I.11)
11 Katsanis T. Fortran program for calculating transonic Thus the continuity equation (I.1) is written:
velocities on a blade to blade surface of a turbomachine.
NASA Report No TND-5427 1970 (pu)x + ( p v ) v ----(/~Up, A$)= + (/.tvpsAS)y (I.12)
12 Camus J-J. Transonic cascade tunnel commissioning tests 2 or:
RD profile. Cambridge University Engineering Department
CUED~A-Turbo~ TR- l O3 1979 (#u), + (#v)v = 0 (I.13)
with
= p -pp, As (I.14)

Int. J. Heat & Fluid Flow 237